A Graph-Based Representation Method for Fashion Color

Round 1

Reviewer 1 Report

The manuscript subject “A Graph Based Representation Method for Fashion Color” is interesting, and the manuscript developed well. However, some minor improvements are as follows:

·       Adding more and newer references could improve the paper;

·       Adding references to the Figures seems necessary;

·       The methodology section needs to be improved.

·       A flowchart of the methods could improve the manuscript;

·       Compared with other methods such as OpenCV, the advantages and disadvantages of the presented method need to be explained.

·       In sections 3 and 4, how are specific objects (cloths) detected in an image/databank?

 

·       The conclusion needs to be improved. It is written very briefly without any information.

Author Response

The manuscript subject “A Graph Based Representation Method for Fashion Color” is interesting, and the manuscript developed well. However, some minor improvements are as follows:

We’d like to thank the reviewer for your time and effort. Indeed, our topic is novel and interesting, which is also fundamental to the area of color trend prediction for fashion apparels. Listed are the replies to your comments. The corresponding changes are made into our manuscript and marked with blue color.

1. Adding more and newer references could improve the paper.

In the process of revision, we have added in more relative references, including reference [7] and reference [21], both addressing the importance of color information in fashion trend prediction and the introduction of machine learning based methods. We’d like to thank the reviewer for suggesting new references if you have any.

2. Adding references to the Figures seems necessary.

Thank you very much for your suggestion, we have quoted the source of the pictures involved in the figures in our revised version. Specifically, the figures from other sources are quoted with references [8], [23] and [24], respectively.

3. The methodology section needs to be improved.

We have improved the methodology section with more detailed theoretical presentation. In addition, we have also added in a flowchart as suggested by the reviewer in the later comments with the corresponding explanation of it. This provides a better presentation of our method and may help the reader to easily understand our approach.

4. A flowchart of the methods could improve the manuscript.

Thanks to your suggestion, we have added the flow chart of the process of building the color graph of a fashion apparel into the methodology section in the revised version. The corresponding description and explanation are also added.

5. Compared with other methods such as OpenCV, the advantages and disadvantages of the presented method need to be explained.

OpenCV is one of the processing libraries that is required in our proposed method, e.g., extracting the pixels from the images, clustering the pixels into different color categories, the counting of the pixels and calculation of the ratio of the colors. Although OpenCV provides powerful functions to capture the detailed information at the granularity of pixel, it does not provide the way to capture the geographical location of the colors as well as constructing the color graph matrices. The corresponding discussions are added into the experimental part in our revised version.

6. In sections 3 and 4, how are specific objects (cloths) detected in an image/databank?

This paper is mainly focusing on the conceptual definition of the color graph and the effectiveness of it on fundamental color research tasks. So, the images we have used are with white background. This is also on purpose to reduce the inaccuracy caused by the currently object segmentation algorithms. In the future work, we will adopt more accurate cloth segmentation methods together with optimized clustering methods.

7. The conclusion needs to be improved. It is written very briefly without any information.

Yes, we agree with the reviewer that the conclusion should be further revised and enriched, we have revised it with more concrete and detailed descriptions in the revised version.

Reviewer 2 Report

The article deals with fashion color research that tries to use the color information of the fashion apparel for style categorization or trend prediction. The major contribution of the article is the proof of the fact that not only the attributes of each color itself but also the relationship of the colors is important and the authors prove the validity of their assertion by a novel knowledge graph-based representation method that not only captures the individual colors but also abstracts the spatial relation of all the colors appear in a single fashion apparel. The authors describe the abstraction of the relationship of colors, the construction of the graph and various aspects of managing their matrix. Moreover, case studies on their proposed approach depict its effectiveness.

 

The article deals with an interesting problem though not so known in the algorithmic community, and the outcome of the work seems to be worth for further investigation. On the other hand, the technical and algorithmic material in the specific paper seems to be limited making it difficult to judge it. According to my opinion more algorithmic material is needed in order for the specific paper to be accepted. For example, the authors should try to discuss (wither theoretically or experimentally) applications of machine learning algorithms in their approach  such as Graph Neural Networks (GNN) or Graph Convolutional Networks (GCN). Hence I vote for major revision.

Author Response

Thank you very much for your valuable and pertinent advice. We agree that the theoretical deduction in our current submission is limited since it is a very initial work that adopts the knowledge graph to represent the color information for a single fashion apparel. Also, thanks to your suggestion, we plan to add in a graph similarity experiment to further verify the effectiveness of our method. Also, we will add in the theoretical relation between our approach and GNN/GCN based approaches in the revised version.

Round 2

Reviewer 2 Report

The article is in a much better shape and I vote for acceptance.